SUPERVISED MACHINE LEARNING APPLICATION FOR ESTIMATION OF SHEAR WAVE AND CHARACTERIZATION OF LOWER GORU FORMATION, SAWAN AREA, LOWER INDUS BASIN.

Abstract

Conventional geophysical techniques applied to vintage seismic and wireline log data have been successful in hydrocarbon exploration. However, the use of machine learning techniques has become increasingly important in overcoming the acquisition based data limitations and subsequently improving the efficiency and accuracy of hydrocarbon exploration. This thesis presents the characterization of the Lower Goru Formation in the Sawan Area of the Lower Indus Basin through the integration of supervised machine learning techniques into geophysical analysis. The primary objectives of this research encompass the precise demarcation of the reservoir of interest, the identification of significant subsurface structures, and an in-depth examination of the tectonic regime in the region. These objectives serve as a foundation for reservoir characterization and hydrocarbon exploration. The methodology employed involves precise tying of horizons to seismic sections, the creation of synthetic seismograms using well data and a process of horizon demarcation on seismic sections. This approach facilitates the generation of spatial time maps for key horizons (D-, C-, and B sands), providing valuable insights into structural variations. Moreover, the petrophysical analysis of well Sawan-01 reveals key reservoir parameters, including volume of shale (20%), porosity (17.3%), effective porosity (11.4%), water saturation (34%), and hydrocarbon content (66%). A comparative analysis with well Sawan-07 highlights the reservoir heterogeneity within the study area. The core of this research lies in the prediction of the DT4S log in well Sawan-01 through a supervised machine learning technique. Gradient Boost Regressor gives 94% accurate prediction for data trained and tested on Wells Sawan-07 and Sawan-08. The model was applied to well Sawan-01, demonstrating its efficacy in estimating critical subsurface properties. Notably, this research concludes with the successful demarcation of hydrocarbon zones through cross-plot analysis utilizing the predicted shear wave data in conjunction with other elastic parameters. The use of predicted log for such analysis is effective for reservoir evaluation, particularly in scenarios where shear wave velocity data is historically lacking in older wells.